2 #ifdef CONFIG_SCHEDSTATS
4 * bump this up when changing the output format or the meaning of an existing
5 * format, so that tools can adapt (or abort)
7 #define SCHEDSTAT_VERSION 14
9 static int show_schedstat(struct seq_file *seq, void *v)
13 seq_printf(seq, "version %d\n", SCHEDSTAT_VERSION);
14 seq_printf(seq, "timestamp %lu\n", jiffies);
15 for_each_online_cpu(cpu) {
16 struct rq *rq = cpu_rq(cpu);
18 struct sched_domain *sd;
22 /* runqueue-specific stats */
24 "cpu%d %u %u %u %u %u %u %u %u %u %llu %llu %lu",
25 cpu, rq->yld_both_empty,
26 rq->yld_act_empty, rq->yld_exp_empty, rq->yld_count,
27 rq->sched_switch, rq->sched_count, rq->sched_goidle,
28 rq->ttwu_count, rq->ttwu_local,
29 rq->rq_sched_info.cpu_time,
30 rq->rq_sched_info.run_delay, rq->rq_sched_info.pcount);
32 seq_printf(seq, "\n");
35 /* domain-specific stats */
37 for_each_domain(cpu, sd) {
38 enum cpu_idle_type itype;
39 char mask_str[NR_CPUS];
41 cpumask_scnprintf(mask_str, NR_CPUS, sd->span);
42 seq_printf(seq, "domain%d %s", dcount++, mask_str);
43 for (itype = CPU_IDLE; itype < CPU_MAX_IDLE_TYPES;
45 seq_printf(seq, " %u %u %u %u %u %u %u %u",
47 sd->lb_balanced[itype],
49 sd->lb_imbalance[itype],
51 sd->lb_hot_gained[itype],
52 sd->lb_nobusyq[itype],
53 sd->lb_nobusyg[itype]);
55 seq_printf(seq, " %u %u %u %u %u %u %u %u %u %u %u %u\n",
56 sd->alb_count, sd->alb_failed, sd->alb_pushed,
57 sd->sbe_count, sd->sbe_balanced, sd->sbe_pushed,
58 sd->sbf_count, sd->sbf_balanced, sd->sbf_pushed,
59 sd->ttwu_wake_remote, sd->ttwu_move_affine,
60 sd->ttwu_move_balance);
68 static int schedstat_open(struct inode *inode, struct file *file)
70 unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32);
71 char *buf = kmalloc(size, GFP_KERNEL);
77 res = single_open(file, show_schedstat, NULL);
79 m = file->private_data;
87 const struct file_operations proc_schedstat_operations = {
88 .open = schedstat_open,
91 .release = single_release,
95 * Expects runqueue lock to be held for atomicity of update
98 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
101 rq->rq_sched_info.run_delay += delta;
102 rq->rq_sched_info.pcount++;
107 * Expects runqueue lock to be held for atomicity of update
110 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
113 rq->rq_sched_info.cpu_time += delta;
115 # define schedstat_inc(rq, field) do { (rq)->field++; } while (0)
116 # define schedstat_add(rq, field, amt) do { (rq)->field += (amt); } while (0)
117 # define schedstat_set(var, val) do { var = (val); } while (0)
118 #else /* !CONFIG_SCHEDSTATS */
120 rq_sched_info_arrive(struct rq *rq, unsigned long long delta)
123 rq_sched_info_depart(struct rq *rq, unsigned long long delta)
125 # define schedstat_inc(rq, field) do { } while (0)
126 # define schedstat_add(rq, field, amt) do { } while (0)
127 # define schedstat_set(var, val) do { } while (0)
130 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
132 * Called when a process is dequeued from the active array and given
133 * the cpu. We should note that with the exception of interactive
134 * tasks, the expired queue will become the active queue after the active
135 * queue is empty, without explicitly dequeuing and requeuing tasks in the
136 * expired queue. (Interactive tasks may be requeued directly to the
137 * active queue, thus delaying tasks in the expired queue from running;
138 * see scheduler_tick()).
140 * This function is only called from sched_info_arrive(), rather than
141 * dequeue_task(). Even though a task may be queued and dequeued multiple
142 * times as it is shuffled about, we're really interested in knowing how
143 * long it was from the *first* time it was queued to the time that it
146 static inline void sched_info_dequeued(struct task_struct *t)
148 t->sched_info.last_queued = 0;
152 * Called when a task finally hits the cpu. We can now calculate how
153 * long it was waiting to run. We also note when it began so that we
154 * can keep stats on how long its timeslice is.
156 static void sched_info_arrive(struct task_struct *t)
158 unsigned long long now = task_rq(t)->clock, delta = 0;
160 if (t->sched_info.last_queued)
161 delta = now - t->sched_info.last_queued;
162 sched_info_dequeued(t);
163 t->sched_info.run_delay += delta;
164 t->sched_info.last_arrival = now;
165 t->sched_info.pcount++;
167 rq_sched_info_arrive(task_rq(t), delta);
171 * Called when a process is queued into either the active or expired
172 * array. The time is noted and later used to determine how long we
173 * had to wait for us to reach the cpu. Since the expired queue will
174 * become the active queue after active queue is empty, without dequeuing
175 * and requeuing any tasks, we are interested in queuing to either. It
176 * is unusual but not impossible for tasks to be dequeued and immediately
177 * requeued in the same or another array: this can happen in sched_yield(),
178 * set_user_nice(), and even load_balance() as it moves tasks from runqueue
181 * This function is only called from enqueue_task(), but also only updates
182 * the timestamp if it is already not set. It's assumed that
183 * sched_info_dequeued() will clear that stamp when appropriate.
185 static inline void sched_info_queued(struct task_struct *t)
187 if (unlikely(sched_info_on()))
188 if (!t->sched_info.last_queued)
189 t->sched_info.last_queued = task_rq(t)->clock;
193 * Called when a process ceases being the active-running process, either
194 * voluntarily or involuntarily. Now we can calculate how long we ran.
196 static inline void sched_info_depart(struct task_struct *t)
198 unsigned long long delta = task_rq(t)->clock -
199 t->sched_info.last_arrival;
201 t->sched_info.cpu_time += delta;
202 rq_sched_info_depart(task_rq(t), delta);
206 * Called when tasks are switched involuntarily due, typically, to expiring
207 * their time slice. (This may also be called when switching to or from
208 * the idle task.) We are only called when prev != next.
211 __sched_info_switch(struct task_struct *prev, struct task_struct *next)
213 struct rq *rq = task_rq(prev);
216 * prev now departs the cpu. It's not interesting to record
217 * stats about how efficient we were at scheduling the idle
220 if (prev != rq->idle)
221 sched_info_depart(prev);
223 if (next != rq->idle)
224 sched_info_arrive(next);
227 sched_info_switch(struct task_struct *prev, struct task_struct *next)
229 if (unlikely(sched_info_on()))
230 __sched_info_switch(prev, next);
233 #define sched_info_queued(t) do { } while (0)
234 #define sched_info_switch(t, next) do { } while (0)
235 #endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */